466 The Feynman path integral
∫t
0
ds
[
m
2
(
dx
ds
) 2
−U(x(s))
]
=
∫−iβh ̄
0
d(−iτ)
[
m
2
(
dx
d(−iτ)
) 2
−U(x(−iτ))
]
=i
∫β ̄h
0
dτ
[
m
2
(
dx
dτ
) 2
+U(x(τ))
]
≡iS[x]. (12.4.11)
Note that the actionS[x] is now the action for paths inimaginary timeτthat start at
x(0) =xand end atx(τ) =x′in imaginary timeτ=β ̄h. The imaginary-time action
S[x] differs from the real-time actionA[x] by the sign of the potential. The actionS[x]
is often called theEuclidean action. In terms of imaginary-time paths, we may write
the density matrix elements as
ρ(x,x′;β) =
∫x(β ̄h)=x′
x(0)=x
Dx(τ) exp
{
−
1
̄h
∫βh ̄
0
dτ
[
1
2
mx ̇^2 (τ) +U(x(τ))
]}
=
∫x(β ̄h)=x′
x(0)=x
Dx(τ) exp
{
−
1
̄h
∫βh ̄
0
dτΛ(x(τ),x ̇(τ))
}
=
∫x′
x
Dxe−S[x]/ ̄h. (12.4.12)
The quantity Λ(x,x ̇) = (m/2) ̇x^2 +U(x) is called theimaginary-time Lagrangianor
Euclidean Lagrangian. The density matrix is constructed by integrating over all paths
x(τ) that satisfyx(0) =x,x(β ̄h) =x′weighted by exp(−S[x]/ ̄h). Since this weight
factor is positive definite, we can find the most important contributions to the func-
tional integral in eqn. (12.4.12) by minimizing the Euclidean action with respect to
the pathx(τ). As we did for the propagator, we consider a pathx(τ) and a nearby
path ̃x(τ) =x(τ) +δx(τ). Ifx(0) =xandx(β ̄h) =x′, thenδx(0) =δx(β ̄h) = 0.
We require that the variationδS=S[x+δx]−S[x] vanish to first order in the path
variationδx. Following the procedure in Section 1.8, the resulting equation of motion
will be exactly the form of the Euler–Lagrange equation applied to Λ:
d
dτ
(
∂Λ
∂x ̇(τ)
)
−
∂Λ
∂x(τ)
= 0. (12.4.13)
However, when we apply the Euler–Lagrange equation to the Euclidean Lagrangian,
we obtain an equation of motion of the form
m
d^2 x
dτ^2
=
∂U
∂x
. (12.4.14)
Eqn. (12.4.14) resembles Newton’s second law except that the force is calculated using
not the potentialU(x) but an inverted potential surface−U(x). This result is not
unexpected: If we transform Newton’s second law in real timemd^2 x/ds^2 =−∂U/∂x